Abstract: 【Objective】Biomass, an important parameter to characterize vegetation activities, is of great significance for plant growth monitoring and yield forecasting. Hyperspectral remote sensing technology based on the unmanned aerial vehicle (UAV) has the advantages of flexibility, non-destructive and wide coverage, and could also timely and accurately estimate vegetation biomass, so it has become one attention topic in remote sensing application. Since saturation problem existed in the inversion of biomass by spectral features, the objective of this study was to propose a 'image and spectrum' fusion index by integrating the biomass-related texture feature into vegetation index.【Method】In this study, the extracted spectral indices and texture features from UAV hyperspectral imagery were used to first construct biomass models, respectively. Secondly, the 'image and spectrum' fusion indices by combining (multiplying or dividing) the biomass-sensitive vegetation index and texture feature were established to solve the saturation problem by spectral information and texture information fusion and to construct biomass model. Finally, the estimation effect of the biomass model constructed by different indices was compared, and then analyze the ability of the 'image and spectrum' fusion indices to estimate biomass. 【Result】 (1) The vegetation index was almost saturated when LAI was no larger than 5, while these 'image and spectrum' fusion indices, VI×sm₆₅₈, VI/ent₆₅₈, VI/dis₆₅₈, VI/con₆₅₈, VI/dis₅₁₄, VI/con₅₁₄, VI/var₅₁₄, VI×con₈₀₂, VI×dis₈₀₂, began to perform saturation when at LAI was larger than 5. Compared with the vegetation index, the anti-saturation ability of the 'image and spectrum' fusion index was improved obviously. (2) Compared with the vegetation index (excepting for GNDVI、NDVI), the anti-saturation ability of the 'image and spectrum' fusion indices (VI×sm₆₅₈, VI/ent₆₅₈, VI/dis₆₅₈, VI/con₆₅₈, VI/dis₅₁₄, VI/con₅₁₄, VI/var₅₁₄, VI×con₈₀₂, VI×dis₈₀₂) improved effectively, and their correlations with biomass improved as well. Meanwhile the biomass model based on the 'image and spectrum' fusion indices performed well, with R² and RMSE values of 0.81 and 826.02 kg·hm^-2, respectively. (3) Compared with spectral index and texture feature, biomass model accuracy by 'image and spectrum' fusion index (R²=0.81) was significantly higher than that of the vegetation index (R² = 0.69) and texture feature (R² = 0.71).【Conclusion】Results showed that both the anti-saturation ability and the accuracy of biomass model constructed by the 'image and spectrum' fusion index were effectively improved, which indicated that spectral information and texture information fusion could achieve a great estimation of winter wheat biomass. The research provided a new way for quantitative inversion of biomass.

Key words: biomass, 'image and spectrum' fusion index, texture feature, saturation, winter wheat